AI-assisted newsroom workflow for event tracking, evidence organization, approval chains, correction handling, and post-publication monitoring.

NewsFlow is not just an AI writing tool. It is a newsroom workflow system designed to support the middle of the reporting process: lead intake, event clustering, evidence structuring, claim review, risk checks, sign-off decisions, correction handling, and post-publication follow-up.

中文简介：NewsFlow 不是一个只围绕“自动写稿”的 AI 工具。它更关注新闻生产中真正高成本、也更高责任的中间流程：线索进入、事件归并、证据组织、Claim 结构化、风险暴露、签发审批、勘误流转，以及发布后的监测与知识回写。

Most AI products for journalism focus on output: drafting, rewriting, summarizing, or style transfer. NewsFlow focuses on a different problem: the hidden coordination and judgment costs that sit before and after the final text.

It is built around a simple belief:

• the hardest part of responsible news work is often not writing itself
• the real cost sits in organizing evidence, judgment, risk, and handoff across roles
• if AI enters this space, it should make the process more visible and reviewable, not less accountable

• self-hosted product exploration
• newsroom workflow demos and teaching
• full-stack prototype extension
• experiments around human-in-the-loop review, approval traceability, and editorial governance

• End-to-end frontend and backend are in place
• Core objects, workflow stages, and key review pages are implemented
• Suitable for self-hosting, demos, and secondary development
• Still closer to a high-fidelity prototype than a production-ready newsroom platform

• CONTRIBUTING.md — contribution guide and development expectations
• SECURITY.md — responsible disclosure and sensitive-data boundaries
• CODE_OF_CONDUCT.md — collaboration expectations for contributors
• docs/OPEN_SOURCE_GUIDE.md — GitHub positioning, topics, and publishing checklist
• docs/GITHUB_ABOUT.md — ready-to-use repository description and About copy
• docs/RELEASE_v0.1.0.md — first public release notes and announcement copy
• docs/SELF_HOSTING.md — local and self-hosted setup guide
• backend/.env.example — backend environment example
• frontend/.env.example — frontend environment example

newsflow/
├── backend/                     # FastAPI backend
│   ├── app/
│   │   ├── agents/             # AI agents by workflow stage
│   │   ├── api/v1/             # REST API endpoints
│   │   ├── core/               # config, database, security, state machine
│   │   ├── models/             # ORM models
│   │   └── services/           # business services
│   ├── migrations/             # Alembic migrations
│   └── tests/                  # backend tests
├── frontend/                    # React + TypeScript frontend
│   └── src/
│       ├── pages/
│       ├── layouts/
│       ├── services/
│       └── stores/
├── docs/                        # open-source and audit documentation
└── image/                       # icon and preview assets

• RSS, API, and manual intake
• deduplication and event clustering
• automatic risk grading
• editor-facing triage suggestions

• 5W1H extraction
• Claim Card generation
• evidence verification
• relationship mapping

• multi-stage redaction gates
• PII detection
• source-protection boundaries
• traceable audit logs

• AI-assisted drafting
• channel adaptation
• semantic drift checks
• platform rule checks

• state-machine-driven workflow
• multi-stage sign-off
• SLA awareness
• explicit human gates

• Python 3.12 or 3.13 recommended for backend setup
• Node.js 20.19+ recommended for frontend setup
• Docker and Docker Compose for local infrastructure

Note: Python 3.14 is not yet recommended here because some native dependencies such as orjson may fail to build depending on your environment.

docker compose up -d

• copy backend/.env.example to backend/.env
• copy frontend/.env.example to frontend/.env.local

cd backend
pip install -r requirements.txt
alembic upgrade head
uvicorn app.main:app --reload --port 8000

cd frontend
npm install
npm run dev

• frontend: http://localhost:5173
• health check: http://localhost:8000/health
• API docs: http://localhost:8000/docs

• frontend uses VITE_API_BASE_URL to connect to the backend
• login warmup/fallback can also use VITE_BACKEND_DIRECT_URL and VITE_BACKEND_HEALTH_URL
• default local API URL is http://localhost:8000/api/v1
• set OPENAI_API_KEY in backend/.env if you want to enable LLM-backed features

• EventCase
• StoryPacket
• ClaimCard
• EvidencePack
• DraftVersion
• ChannelPackage
• ReviewBundle
• ApprovalTask
• DecisionLog
• SourceVault
• RiskReport
• CorrectionTicket
• AuditLog

• FastAPI
• SQLAlchemy + Alembic
• aiosqlite / asyncpg
• LangChain Core
• Gunicorn + Uvicorn

• React 18
• TypeScript
• Ant Design 5
• React Router 6
• Vite
• Zustand

• OpenAI GPT-4o / GPT-4o-mini
• LangChain OpenAI

• core object model and state machine
• multi-stage agents across input, cognition, governance, and production
• sign-off center and human override support
• post-publication monitoring and correction loop
• open-source packaging and self-hosting setup
• more complete self-hosting docs
• finer-grained evidence anchors and citation mapping
• stronger permissions, metrics, and long-term memory governance

• docs/AUDIT_FINDINGS.md — audit findings across backend, frontend, and integration
• docs/OPEN_SOURCE_GUIDE.md — open-source packaging notes
• docs/GITHUB_ABOUT.md — repository description and public-facing copy
• docs/RELEASE_v0.1.0.md — first public release notes
• docs/GITHUB_PAGE_SETUP.md — final GitHub page copy and launch checklist
• docs/PUBLICATION_AUDIT.md — public-release scope and sensitive-info audit
• docs/SELF_HOSTING.md — self-hosting and deployment guidance

If you want this repository to feel complete on GitHub, the next good steps are:

• upload image/newsflow-social-preview.png as the GitHub social preview image
• copy the text from docs/GITHUB_ABOUT.md into the repository About section
• publish a first Release using docs/RELEASE_v0.1.0.md
• follow docs/REPO_MAINTENANCE_CHECKLIST.md for branch protection, issue hygiene, and release upkeep

This project is released under the MIT License.